Molten steel temperature drop amount prediction method in steel making process and molten steel temperature drop amount prediction device

Abstract

To provide a molten steel temperature drop amount prediction technique in the steel making process with improved temperature drop prediction accuracy.SOLUTION: Information including the number of times a molten steel ladle has been used in the past to receive molten steel of a certain charge from a converter process, empty ladle waiting time, molten steel ladle transport time, processing time, molten steel temperature at completion, processing route, and the number of times the tundish is continuously used for receiving the molten steel of the charge in the continuous casting process are used as teacher data to pre-construct a temperature drop amount prediction model that predicts the molten steel temperature drop amount during the period from the end of the current process to the subsequent process, information including the target temperature in the subsequent process of the molten steel of the charge for which the temperature drop is predicted, the number of times the molten steel ladle is used, the scheduled empty ladle waiting time, scheduled transport time, scheduled processing time, scheduled molten steel temperature at completion, scheduled processing route, and the scheduled number of times the tundish is continuously used for receiving the molten steel of the target charge in the continuous casting process are inputted to the temperature drop amount prediction model, then, the molten steel temperature drop amount during the period from the end of the current process of the target charge to the subsequent process is obtained.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to calculating the amount of drop in molten steel temperature in a steelmaking process. [Background technology]

[0002] The steelmaking process, which starts from the converter furnace and ends with the continuous casting process, can be divided into two stages: Case 1, where the converter furnace is followed by a secondary refining process (including a vacuum degassing process) before the continuous casting process, and Case 2, where the converter furnace is followed directly by the continuous casting process. A key challenge in the steelmaking process, particularly the continuous casting process, is the high temperature of the molten steel. High temperatures increase the risk of remelting breakouts, necessitating a slower casting speed in the continuous casting process. Furthermore, high tapping temperatures increase costs due to the increased lime consumption in the converter furnace. Therefore, lowering the tapping temperature and maintaining a low molten steel temperature are essential for cost reduction and trouble prevention. However, excessively low molten steel temperatures can lead to operational issues, such as clogging of the tundish sliding nozzle in the continuous casting process. Therefore, accurately calculating the amount of molten steel temperature reduction in the steelmaking process is a key challenge.

[0003] One method for accurately calculating the amount of drop in molten steel temperature is to determine a weighting factor for the conditions of the molten steel ladle and transportation time in the steelmaking process, and then calculate the amount of temperature drop.

[0004] For example, Patent Document 1 discloses a technique for automatically determining the temperature of molten steel in a process based on a request from a downstream process using a hierarchical neural network. This technique first stores actual data on the number of uses of a molten steel ladle, standby time, transport time, number of uses of a tundish, and the amount of alloy added to the molten steel in the neural network. Then, weighting coefficients and thresholds for the neural network are determined. The number of uses of a molten steel ladle, standby time, planned transport time, number of uses of a tundish, and planned amount of alloy added to the molten steel for the charge to be temperature-adjusted are input to the input layer of the neural network, and the temperature drop of the molten steel is calculated. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 08-003621 Summary of the Invention [Problem to be solved by the invention]

[0006] However, the above-mentioned conventional techniques have the following problems to be solved. That is, the method described in Patent Document 1 does not fully consider the operating conditions in the secondary refining process, and there is a problem in that the accuracy of estimating the temperature drop is insufficient, as shown in the following (1) to (4). (1) The higher the end temperature of the secondary refining process, the greater the temperature drop until the continuous casting process. Patent Document 1 does not take into account the end temperature of the secondary refining process in the "charge subject to temperature adjustment," and therefore cannot fully evaluate this effect in the charge subject to temperature adjustment. (2) The longer the processing time in the secondary refining process, the smaller the temperature drop before the continuous casting process tends to be. This is because the convection phenomenon of molten steel in the ladle changes with the processing time. (3) The amount of temperature drop before the continuous casting process varies depending on whether or not a tundish weir is installed to control the molten steel flow. This is because the amount of heat removal from the molten steel changes depending on the tundish weir, and the convection phenomenon of the molten steel changes due to the change in the molten steel flow. (4) The temperature drop during the continuous casting process varies depending on the thickness and width of the billet. This is because the temperature drop during the continuous casting process varies depending on the casting volume per unit time.

[0007] The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to provide a technology for predicting the amount of temperature drop of molten steel in a steelmaking process, which improves the accuracy of temperature drop prediction by taking into account the operating conditions in the secondary refining process and the continuous casting process. [Means for solving the problem]

[0008] The present invention advantageously solves the above-mentioned problems by providing a method for predicting the amount of temperature drop of molten steel in a steelmaking process, in which molten steel is received in a molten steel ladle from a converter process, passes through a secondary refining process, and is finally transported to a continuous casting process for processing, where the current process is a converter process or a secondary refining process, and the corresponding subsequent process is a secondary refining process or a continuous casting process, respectively, and the method predicts the amount of temperature drop of molten steel in a molten steel ladle that has received a certain charge in the past, the waiting time of an empty ladle after discharging the molten steel received previously until the molten steel of the charge is received, the transport time of the molten steel ladle from the current process to the subsequent process, the processing time in the secondary refining process, the temperature of the molten steel at the end of the secondary refining process, the processing path in the secondary refining process, and the size of a tundish that receives the molten steel of the charge in the continuous casting process. The method is characterized in that a temperature drop prediction model is constructed in advance using information including the number of consecutive uses as training data to predict the amount of temperature drop of molten steel from the end of the current process to the subsequent process, and information including the target temperature of the molten steel of the target charge in the subsequent process for the temperature drop prediction, the number of uses of the molten steel ladle to receive the molten steel of the target charge, the planned empty ladle waiting time of the molten steel ladle, the planned transportation time from the current process to the subsequent process, the planned processing time in the secondary refining process, the planned molten steel temperature at the end of the secondary refining process, the planned processing path in the secondary refining process, and the planned number of consecutive uses of the tundish to receive the molten steel of the target charge in the continuous casting process is input into the temperature drop prediction model to determine the amount of temperature drop of the molten steel from the end of the current process to the subsequent process.

[0009] The method for predicting the temperature drop of molten steel in a steelmaking process according to the present invention includes the steps of: (a) the teacher data further includes information on the thickness and width of the slab in the continuous casting process of the past charge and on the presence or absence of a tundish weir for controlling the flow of molten steel in the tundish receiving the charge, and when calculating the amount of temperature drop of molten steel from the end of the current process of the target charge to the subsequent process, the information to be input into the temperature drop prediction model further includes the thickness and width of the slab in the continuous casting process of the target charge and on the presence or absence of a tundish weir for controlling the flow of molten steel in the tundish receiving the target charge; (b) The temperature drop prediction model is constructed by machine learning using any one of logistic regression analysis, decision tree, neural network, and deep learning; (c) The temperature drop prediction model is constructed by machine learning using a neural network, and is learned by inputting the training data into the neural network, calculating weighting coefficients in the neural network that stores the training data, converting the calculated weighting coefficients into internal states of neurons, and selecting and validating internal states from the converted internal states. This would be a more preferable solution to the problem.

[0010] The apparatus for predicting a temperature drop of molten steel according to the present invention, which advantageously solves the above-mentioned problems, is an apparatus used in the method for predicting a temperature drop of molten steel in a steelmaking process, and includes a past performance input unit that inputs past performance data, including the number of times the molten steel ladle has been used to receive a certain charge in the past, the waiting time of the empty ladle after discharging the molten steel received previously until the molten steel of the charge is received, the transport time of the molten steel ladle from the current process to the subsequent process, the processing time in the secondary refining process, the molten steel temperature at the end of the secondary refining process, the processing path in the secondary refining process, and the number of consecutive uses of the tundish to receive the molten steel of the charge in the continuous casting process, as training data into a pre-constructed temperature drop prediction model; a learning unit that trains the temperature drop prediction model using the training data; and a temperature drop prediction function that is input to the temperature drop prediction model. the target information input unit inputs information including the target temperature of the molten steel in the subsequent process of the target charge, the number of uses of the molten steel ladle to receive the molten steel of the target charge, the planned empty ladle standby time of the molten steel ladle, the planned transport time from the current process to the subsequent process, the planned processing time in the secondary refining process, the planned molten steel temperature at the end of the secondary refining process, the planned processing path in the secondary refining process, and the planned number of consecutive uses of the tundish to receive the molten steel of the target charge in the continuous casting process; a prediction unit predicts the amount of temperature drop of the molten steel from the end of the current process of the target charge to the subsequent process using the data input in the past performance input unit and the target information input unit, based on the temperature drop amount prediction model learned in the learning unit; and an output unit outputs the predicted amount of temperature drop of the molten steel.

[0011] The molten steel temperature drop prediction device according to the present invention includes: (d) the past performance input unit further inputs into the training data information on the thickness and width of the slab in the continuous casting process of the past charge and on the presence or absence of a tundish weir for controlling the flow of molten steel in the tundish that receives the charge, and the target information input unit further inputs information on the thickness and width of the slab in the continuous casting process of the charge that is the target of the temperature drop prediction and on the presence or absence of a tundish weir for controlling the flow of molten steel in the tundish that receives the target charge, (e) The temperature drop prediction model is constructed by machine learning using any one of logistic regression analysis, decision tree, neural network, and deep learning; (f) the temperature drop prediction model is constructed by machine learning using a neural network; The training data is input to a neural network, weighting coefficients are calculated in the neural network that stores the training data, the weighting coefficients are converted into internal states of neurons based on the calculated weighting coefficients, and an internal state to be enabled is selected from the converted internal states to learn the training data. This would be a more preferable solution to the problem. [Effects of the Invention]

[0012] According to the present invention, the temperature drop of molten steel in the steelmaking process is predicted by taking into account the number of uses of the molten steel ladle, the waiting time of the empty ladle, the transport time of the molten steel ladle from the current process to the downstream process, the processing time and processing path in the secondary refining process, the molten steel temperature at the end of the secondary refining process, and the number of continuous uses of the tundish in the continuous casting process. This makes it possible to evaluate differences in the amount of heat stored in the molten steel ladle, the amount of heat released from the molten steel, and the convection of the molten steel, thereby improving the accuracy of the prediction of the temperature drop of molten steel in the steelmaking process. Furthermore, the temperature drop of molten steel in the steelmaking process is predicted by taking into account the thickness and width of the slab in the continuous casting process and the presence or absence of a tundish weir for controlling the flow of molten steel in the tundish. This also makes it possible to evaluate differences in the amount of heat released from the molten steel due to changes in the cast volume per unit time in the continuous casting process, the heat transfer from the molten steel to the tundish, and the convection of the molten steel in the tundish. This further improves the accuracy of the prediction of the temperature drop of molten steel in the steelmaking process. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a conceptual diagram showing the configuration of a molten steel temperature drop prediction device according to one embodiment of the present invention. [Figure 2]1 is a graph showing the relationship between the temperature of molten steel at the end of a secondary refining process and the rate of decrease in the temperature of molten steel up to a continuous casting process. [Figure 3] 1 is a graph showing the relationship between the reflux time in an RH vacuum refining process and the rate of decrease in molten steel temperature from the end of the secondary refining process to the continuous casting process. [Figure 4] 1 is a graph showing the influence of the presence or absence of a tundish dam on the relationship between the elapsed time from the end of the secondary refining process to the continuous casting process and the amount of temperature drop of molten steel. DETAILED DESCRIPTION OF THE INVENTION

[0014] The following describes in detail the embodiments of the present invention. Note that the following embodiments are merely examples of methods for realizing the technical idea of ​​the present invention, and are not intended to limit the configuration to the following. In other words, the technical idea of ​​the present invention can be modified in various ways within the technical scope described in the claims.

[0015] First Embodiment As an example of an embodiment, a method for predicting a molten steel temperature drop amount will be described below, where a secondary refining process is defined as a current process and a continuous casting process is defined as a subsequent process in a steelmaking process. Fig. 1 is a schematic diagram showing the configuration of a molten steel temperature drop amount prediction device suitable for application to the method for predicting a molten steel temperature drop amount in a steelmaking process according to this embodiment.

[0016] In this embodiment, the temperature drop amount of molten steel is predicted using a temperature drop amount prediction model that is machine-learned in advance using past operational performance data as training data. In the example of Figure 1, a neural network is used as the temperature drop amount prediction model.

[0017] As shown in Fig. 1(a), the molten steel temperature prediction device has a past performance input unit 1A, a learning unit 1B, and an output unit 1C, which construct a temperature drop amount prediction model in advance. As shown in Fig. 1(b), the device has a target information input unit 2A, a prediction unit 2B, and an output unit 2C, which predict the temperature drop amount of a target charge for temperature drop prediction. When a neural network is used for the temperature drop amount prediction model, the past performance input unit 1A and the target information input unit 2A are parameter input units, and the learning unit 1B and the prediction unit 2B are internal information conversion units.

[0018] In this embodiment, first, the past operation results are input to the past operation results input unit 1A, and the following is input for each charge: (P1) Number of times the molten steel ladle is used, (P2) The waiting time for the empty ladle from the time when the previously received molten steel is discharged until the molten steel of the charge is received. (P3) Transport time of the molten steel ladle from the current process to the next process, (P4) Processing time in the secondary refining process, (P5) The temperature of molten steel at the end of the secondary refining process, (P6) Processing route in the secondary refining process, and (P7) The number of consecutive uses of the tundish that receives the molten steel of the charge in the continuous casting process The neural network is constructed using a known machine learning algorithm. In this case, if the internal temperature of the molten steel ladle when it is empty can be measured, it is preferable to simultaneously collect and store the internal temperature as a past record and store it in the neural network.

[0019] The learning unit 1B calculates weighting coefficients using a known method from a neural network that stores the accumulated data (P1) through (P7), preferably including the internal temperature measured during an empty ladle, and converts the weighting coefficients into internal neuron states. The learning unit 1B selects an internal neuron state that is effective for calculating the molten steel temperature drop amount from the end of the current process to the next process. Using the selected internal state, a predicted value for the molten steel temperature drop amount can be output to the output unit 1C. Here, if there is a large discrepancy between the predicted value and the actual value, it is preferable to learn to change the weighting coefficients to reduce the discrepancy. For example, Figure 2 shows a graph of the effect of the molten steel temperature at the end of the secondary refining process on the average molten steel temperature drop rate from the end of the secondary refining process to 40 minutes after the start of continuous casting. It can be seen that the higher the molten steel temperature at the end of the secondary refining process, the greater the molten steel temperature drop rate. Figure 3 also shows a graph of the effect of the reflux time in the RH vacuum refining process on the molten steel temperature drop rate. It can be seen that the longer the reflux time, the smaller the molten steel temperature drop rate.

[0020] Next, in the target information input section 2A, from the operation schedule of the charge for which the temperature drop amount is to be predicted, (A1) The number of times the ladle is used to receive the molten steel of the target charge; (A2) Planned empty ladle waiting time, (A3) Estimated transportation time from the current process to the next process, (A4) Planned processing time in the secondary refining process; (A5) The planned molten steel temperature at the end of the secondary refining process, (A6) Planned processing route in the secondary refining process; (A7) The number of planned consecutive uses of the tundish that receives the molten steel of the target charge in the continuous casting process; and (A8) Target temperature of the molten steel in the subsequent process of the target charge are collected and accumulated, and stored in the neural network. At this time, if it is possible to measure the internal temperature of the molten steel ladle when it is empty, it is preferable to simultaneously collect and accumulate this internal temperature as a past record and store it in the neural network.

[0021] Prediction Unit 2B converts the collected data for the target charge into the internal state of the neuron based on the weighting coefficients learned by Learning Unit 1B. Then, based on the internal state selected as valid by Learning Unit 1B, a predicted value for the amount of molten steel temperature drop from the end of the target charge's current process to the next process is output to Output Unit 2C.

[0022] Table 1 shows the results of actual temperature drop predictions using the temperature drop prediction model constructed and trained as described above. The secondary refining process, which was the RH vacuum refining process, was performed. After the process, the molten steel ladle was transported to a continuous casting machine, and the downstream continuous casting process was carried out. The temperature drop was evaluated from the end of the RH vacuum refining process until 40 minutes after the start of continuous casting. The results in Table 1 show that the accuracy rate of the molten steel temperature drop predicted using this embodiment was 82.7%, an improvement over the conventional method described in Patent Document 1. Furthermore, the miscalculation rate for this embodiment was 1.8%, an improvement of 0.4% over the conventional method. Here, the miscalculation rate refers to the error between the predicted and measured temperature drop, i.e., a prediction error of ±5°C or less is considered a hit, and is expressed as a percentage of the total number of hits. Furthermore, the miscalculation rate refers to a prediction error of ±10°C or more is considered a miss, and is expressed as a percentage of the total number of misses.

[0023] [Table 1]

[0024] Second Embodiment In addition to the information entered in the first embodiment, the past performance input unit 1A inputs information about each charge from past operation performance. (P9) Thickness and width of slab in continuous casting process, and (P10) Installation of tundish weir This information is collected, stored, and stored in a neural network, which is constructed using a known machine learning algorithm.

[0025] As in the first embodiment, the learning unit 1B calculates weighting coefficients using a known method from a neural network that stores the accumulated data (P1) to (P7) as well as the accumulated data (P9) and (P10), preferably the internal temperature measured when the ladle was empty, and converts the weighting coefficients into internal states of the neurons. The learning unit 1B selects an internal state of the neuron that is effective for calculating the amount of molten steel temperature drop from the end of the current process to the subsequent process. Using the selected internal state, a predicted value of the amount of molten steel temperature drop can be output to the output unit 1C. The weighting coefficients are learned in the same way as in the first embodiment. For example, Figure 4 is a graph showing the effect of the presence or absence of a tundish weir on the relationship between the elapsed time from the end of the secondary refining process to the continuous casting process and the amount of molten steel temperature drop.

[0026] Table 2 shows the actual results of temperature drop predictions made using the temperature drop prediction model constructed and trained as described above. The training data and target charges were the same as those in the first embodiment. In this embodiment, the hit rate was further improved and the miss rate was reduced.

[0027] [Table 2]

[0028] <Other embodiments> In the above embodiment, the secondary refining process is an RH vacuum refining process. The secondary refining process can also be similarly applied to, for example, ladle refining using arc heating. When the current process is a converter process, the subsequent process is a secondary refining process, and the current process is further set as a secondary refining process and a continuous casting process, and the molten steel temperature drop amount can be estimated repeatedly. If the estimated molten steel temperature drop indicates that the molten steel temperature in the subsequent process deviates from the target temperature by a predetermined value or more, it is preferable to take necessary measures, such as shortening or extending the processing time of the current process, or heating or cooling the molten steel. Instead of using a neural network, the temperature drop prediction model can also be constructed using machine learning, such as logistic regression analysis, decision trees, or deep learning. [Explanation of symbols]

[0029] 1A Past performance input section 1B Learning Department 1C output section 2A Target information input section 2B Prediction Department 2C output section

Claims

1. In a steelmaking process in which molten steel is received from a converter process into a molten steel ladle, the molten steel passes through a secondary refining process, and is finally transported to a continuous casting process for processing, when the process itself is a converter process or a secondary refining process and the corresponding subsequent process is a secondary refining process or a continuous casting process, a temperature drop prediction model for predicting the amount of temperature drop of molten steel from the end of a process to a subsequent process is constructed in advance using as training data information including the number of times the molten steel ladle has been used to receive a particular charge in the past, the waiting time of the empty ladle after the previous molten steel has been discharged until the molten steel of the current charge is received, the transport time of the molten steel ladle from the current process to the subsequent process, the processing time in the secondary refining process, the molten steel temperature at the end of the secondary refining process, the processing path in the secondary refining process, and the number of consecutive uses of a tundish that receives the molten steel of the current charge in the continuous casting process; a target temperature of the molten steel in a downstream process for the charge for which temperature drop is to be predicted, the number of uses of the ladle to receive the molten steel of the target charge, the planned waiting time for the ladle, the planned transport time from the current process to the downstream process, the planned processing time in the secondary refining process, the planned molten steel temperature at the end of the secondary refining process, the planned processing path in the secondary refining process, and the planned number of consecutive uses of the tundish to receive the molten steel of the target charge in the continuous casting process, are input into the temperature drop prediction model, and the amount of temperature drop of the molten steel from the end of the current process to the downstream process is calculated.

2. The teaching data further includes information on the thickness and width of a slab in the continuous casting process of the past charge and on the presence or absence of a tundish weir for controlling the flow of molten steel in a tundish receiving the charge, 2. The method for predicting a temperature drop of molten steel in a steelmaking process according to claim 1, wherein, when calculating a temperature drop amount of the molten steel from the end of a process of the target charge to a subsequent process, information to be input into the temperature drop prediction model further includes a slab thickness and a slab width in a continuous casting process of the target charge and whether a tundish weir for controlling the flow of molten steel is installed in a tundish that receives the target charge.

3. 3. The method for predicting a temperature drop amount of molten steel in a steelmaking process according to claim 1, wherein the temperature drop prediction model is constructed by machine learning using any one of logistic regression analysis, decision tree, neural network, and deep learning.

4. the temperature drop prediction model is constructed by machine learning using a neural network, 4. The method for predicting a temperature drop of molten steel in a steelmaking process according to claim 3, wherein the training data is input to a neural network, weighting coefficients are calculated in the neural network that has stored the training data, conversion is performed into internal states of neurons based on the calculated weighting coefficients, and an internal state to be enabled is selected from the converted internal states for learning.

5. 2. An apparatus used in the method for predicting a temperature drop of molten steel in a steelmaking process according to claim 1, a past performance input unit that inputs past performance data, including the number of times the molten steel ladle has been used to receive a particular charge in the past, the waiting time for the empty ladle after the previously received molten steel is discharged until the molten steel for the charge is received, the transport time for the molten steel ladle from its own process to the subsequent process, the processing time in the secondary refining process, the molten steel temperature at the end of the secondary refining process, the processing path in the secondary refining process, and the number of consecutive uses of the tundish for receiving the molten steel for the charge in the continuous casting process, into a pre-constructed temperature drop prediction model as training data; a learning unit that learns the temperature drop amount prediction model using the training data; a target information input unit for inputting information into the temperature drop prediction model, including a target temperature in a downstream process of the molten steel of the target charge for temperature drop prediction, the number of uses of a molten steel ladle to be used to receive the molten steel of the target charge, a planned waiting time for the empty ladle of the molten steel ladle, a planned transportation time from the current process to a downstream process, a planned processing time in the secondary refining process, a planned molten steel temperature at the end of the secondary refining process, a planned processing path in the secondary refining process, and a planned number of consecutive uses of a tundish to receive the molten steel of the target charge in a continuous casting process; a prediction unit that predicts a molten steel temperature drop amount from the end of the target process to a subsequent process using the data inputted in the past performance input unit and the target information input unit and the temperature drop amount prediction model learned in the learning unit; an output unit that outputs the predicted result of the temperature drop of the molten steel; A molten steel temperature drop prediction device comprising:

6. the past performance input unit further adds information on the thickness and width of the slab in the continuous casting process of the past charge and on whether a tundish weir for controlling the flow of molten steel in a tundish receiving the charge is installed and inputs the added information into the training data; the target information input unit additionally inputs a slab thickness and a slab width in a continuous casting process of the target charge for temperature drop prediction, and information on whether a tundish weir for controlling the flow of molten steel is installed in a tundish receiving the target charge. The apparatus for predicting a temperature drop amount of molten steel according to claim 5.

7. 7. The apparatus for predicting a temperature drop amount of molten steel according to claim 5, wherein the temperature drop amount prediction model is constructed by machine learning using any one of logistic regression analysis, decision tree, neural network, and deep learning.

8. the temperature drop prediction model is constructed by machine learning using a neural network, The training data is input to a neural network, weighting coefficients are calculated in the neural network that stores the training data, the weighting coefficients are converted into internal states of neurons based on the calculated weighting coefficients, and an internal state to be enabled is selected from the converted internal states and learned. The apparatus for predicting a temperature drop of molten steel according to claim 7.

Images